{smcl}
{* 20 Feb 2002/19 Jul 2006}{...}
{hline}
help for {hi:winsor}
{hline}

{title:Winsorizing a variable}

{p 8 17 2}
{cmd:winsor}
{it:varname}
[{cmd:if} {it:exp}]
[{cmd:in} {it:range}]
, {cmdab:g:enerate(}{it:newvar}{cmd:)}
{break}{c -(} {cmd:p(}{it:#}{cmd:)} {c |} {cmd:h(}{it:#}{cmd:)} {c )-}
[ {c -(} {cmdab:high:only} {c |} {cmdab:low:only} {c )-} ]


{title:Description}

{p 4 4 2}{cmd:winsor} takes the non-missing values of a variable {it:x}
ordered such that

{p 8 8 2}{it:x_1} <=  ... <= {it:x_n}

{p 4 4 2}and generates a new variable {it:y} identical to {it:x} except that
the {it:h} highest and {it:h} lowest values are replaced by the next value
counting inwards from the extremes:

{p 8 8 2}{it:y_1}, ... , {it:y_h} = {it:y_(h + 1)}

{p 8 8 2}{it:y_n}, ... , {it:y_(n - h + 1)} = {it:y_(n - h)}

{p 4 4 2}{it:h} can be specified directly or indirectly by specifying a
fraction {it:p} of the number of observations {it:n}:

{p 8 8 2}{it:h} = [ {it:p n} ]

{p 4 4 2} where [ ] denotes integer part. This transformation is named after
the biostatistician Charles P. Winsor (1895-1951): see, for example, Tukey
(1962).  For more discussion and references, see Barnett and Lewis (1994).

{p 4 4 2}Charles (Charlie) Winsor was educated at Harvard as an engineer and
then worked for the New England Telephone and Telegraph Company, but his
interests shifted to biological research and biostatistics. After further
study at Johns Hopkins and Harvard, he held posts at Iowa State College and
Johns Hopkins; in between, in the Second World War, he did government work
at Princeton.


{title:Options}

{p 4 8 2}{cmd:generate(}{it:newvar}{cmd:)} specifies the name of the new
variable.  It is a required option.

{p 4 8 2}{cmd:p(}{it:#}{cmd:)} specifies the fraction of the observations to
be modified in each tail. {it:p} should be greater than 0 and less than 0.5
and imply a value of {it:h} as just below.

{p 4 8 2}{cmd:h(}{it:#}{cmd:)} specifies the number of the observations to
be modified in each tail. {it:h} should be at least 1 and less than half the
number of non-missing observations.

{p 4 4 2}Just one of {cmd:p()} and {cmd:h()} should be specified.

{p 4 8 2}{cmd:highonly} and {cmd:lowonly} specify that Winsorizing should be
one-sided, referring only to the tail with the highest values or only to the
tail with the lowest values, respectively. These options should not be
specified together.


{title:Examples}

{p 4 8 2}{inp:. winsor mpg, gen(Wmpg) h(3)}

{p 4 8 2}{inp:. winsor mpg, gen(Wmpg2) p(0.1)}


{title:References}

{p 4 8 2}Anonymous. 1951. In memoriam: Charles P. Winsor.
{it:Biometrics} 7: 221.

{p 4 8 2}Barnett, V. and Lewis, T. 1994. {it:Outliers in statistical data.}
Chichester: John Wiley. [Previous editions 1978, 1984.]

{p 4 8 2}Tukey, J.W. 1962. The future of data analysis.
{it:Annals of Mathematical Statistics} 33: 1{c -}67.


{title:Author}

{p 4 4 2}Nicholas J. Cox, Durham University, U.K.{break}
         n.j.cox@durham.ac.uk


